Statistics from Altmetric.com
The scientific community is currently very interested in the effect modification of air pollution by meteorological variables. The recent paper by Dr Nawrot and colleagues reports findings from a time-series epidemiology study on the effects of temperature on the association between relative mortality risk and PM10 in Belgium.1 Their 7-year data showed a stronger overall association and no threshold between PM10 and mortality in summer, but a weaker association and a threshold of 30 μg/m3 in winter. Based on these findings they state that “the shape of the dose–response relationship curve between air pollution and mortality strongly depends on outdoor temperature”.
We would like to point out that, although a weaker dependency cannot be ruled out, at least a substantial fraction of the observation is most likely attributable to the well-known relationship between personal exposures and air pollution levels at fixed monitoring sites. This relationship is strongly modified by the amount of time spent indoors. Infiltration of outdoor air PM2.5 particles indoors reduces exposure levels relative to ambient levels by 30–40% on average in Europe.2 The associated strong seasonal pattern3 is caused by variation in the times for which windows are kept open, e.g. in Helsinki windows are kept open for a median of 0.3 h/day in winter compared with 24 h/day in summer.4 Ventilation via open windows particularly affects infiltration of the coarse fraction of PM10.
Janssen et al. have shown that in the US an increase in the prevalence of air conditioning in cities results in a decrease in the slope and significance of the daily morbidity–ambient PM10 regression.5 They believe that this is due to the difference in the ambient PM10 penetration between the seasons and the effects of air conditioning versus no air conditioning.
Therefore, while the work by Nawrot and colleagues opens up an important discussion on effect modification by meteorological variables, the most urgent topic in this field is to quantify the effects on exposures. Only after these effects have been controlled can analysis of the effect modification by temperature and other variables be properly conducted.
What this paper adds
The scientific community, especially epidemiologists, are very excited about the topic of effect modification, by temperature and other weather variables, of the health impact of air pollution, especially particulate matter. However, at least to a substantial degree, this modification is caused by seasonal variations in population behaviour, and not necessarily by any differences in the toxicity of the particles. Therefore, exposure modification should be studied and included in the interpretation of findings such as those reported by Nawrot et al.
The toxicity of different particles is of the utmost importance in exposure reduction policy development. Furthermore, estimates of the public health costs associated with air pollution are dependent on the dose–response functions estimated in epidemiological studies. If interpreted incorrectly, very wrong actions – or no action at all – may be taken at the expense of public health.
Hänninen and Jantunen argued that an analysis of effect modification by temperature can be properly conducted only after the effects of meteorology on exposure have been thoroughly quantified.
Interaction or effect modification refers to the extent to which the joint effect of two risk factors on disease or mortality differs from the independent effect of each of the factors. In essence, heterogeneous data should not be pooled and homogeneity and linearity are two assumptions that should be tested in a well-conducted statistical analysis. In our study,1 the association between mortality and daily concentrations of fine particulate air pollution was not linear at lower temperatures. Therefore, using all data would have biased the overall estimate. Moreover, there was a clear rationale for interaction testing because the literature shows that both high and low temperatures increase mortality2–4 and that air pollution is associated with temperature.5
With our analysis, we can only speculate about the mechanisms underlying the much stronger association found between mortality and PM10 during warmer periods, even though the PM10 levels reach higher values in winter. As we discussed in our paper, three explanations can be proposed. First, as postulated by Hänninen and Jantunen, the higher relative effects during the summer might be a consequence of spending more time outdoors or because of indoor and outdoor PM10 are more similar in the summer. Hänninen and Jantunen argue that the strong seasonal pattern may be influenced by the times windows are kept open, with median times in Helsinki of 0.3 h/day in winter compared with 24 h/day in summer. Such large differences, however, probably do not apply in Belgium, which has a temperate climate, with a mean average temperature of 2.6°C in January. Two other potential explanations, which we discussed in our paper, include the lower background mortality in summer, resulting in a larger pool of susceptible people in summer, and the component-specific toxicity of PM10, which may differ across the temperature range. The discussion on the interaction between meteorology and particulate air pollution on acute health effects in terms of triggering mortality will continue for some time until appropriate experimental studies in animals or human volunteers have been completed. An experimental study that exposed isolated rat macrophages to ambient particulate matter collected during winter, spring and summer (in Amsterdam, Lodz, Oslo and Rome) showed that PM10 samples collected in summer were more potent at inducing inflammatory cytokines.6
Since exposure estimates are also based on assumptions, with many uncertainties, there is an urgent need for simple valid biomarkers of internal particle exposure. Carbon load in lung macrophages7 might represent such a measure to test the underlying reason(s) for the larger health effects in the summer than in winter. However, whatever the uncertainty of exposure measures might be, this is not by itself a reason to ignore heterogeneity and effect modification within data.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.